Written by Chengyang Li based on the original demo code from Soonmin Hwang.

See demo_test.m for an example.

Note: We provide evalutaion results in terms of log-average miss rate as well as recall on all 9 different settings, i.e. Reasonable-all, Reasonable-day, Reasonable-night, Scale=near, Scale=medium, Scale=far, Occ=none, Occ=partial, Occ=heavy (see CVPR15 paper for details). Both using the origninal and the improved (in parentheses) test annotaions. However we strongly suggest you report results using the improved test annotaions only to enable a fair comparision.